app.py

from gradio.outputs import Label
from icevision.all import *
from icevision.models.checkpoint import *
import PIL
import gradio as gr
import os

# Load model
checkpoint_path = "model_checkpoint.pth"
checkpoint_and_model = model_from_checkpoint(checkpoint_path)
model = checkpoint_and_model["model"]
model_type = checkpoint_and_model["model_type"]
class_map = checkpoint_and_model["class_map"]

# Transforms
img_size = checkpoint_and_model["img_size"]
valid_tfms = tfms.A.Adapter([*tfms.A.resize_and_pad(img_size), tfms.A.Normalize()])

# Populate examples in Gradio interface
examples = [
    ['1.jpg'],
    ['2.jpg'],
    ['3.jpg']
]

def show_preds(input_image):
    img = PIL.Image.fromarray(input_image, "RGB")
    pred_dict = model_type.end2end_detect(img, valid_tfms, model, 
                                          class_map=class_map, 
                                          detection_threshold=0.5,
                                          display_label=False, 
                                          display_bbox=True, 
                                          return_img=True, 
                                          font_size=16, 
                                          label_color="#FF59D6")
    return pred_dict["img"]

gr_interface = gr.Interface(
    fn=show_preds,
    inputs=["image"],
    outputs=[gr.outputs.Image(type="pil", label="RetinaNet Inference")],
    title="Aircraft Detector",
    examples=examples,
)
gr_interface.launch(inline=False, share=False, debug=True)